joseph.ergo@proton.me | Portfolio | Resume PDF | Linked-In | +212 713-617-633

Available immediately for full/part-time remote roles

FREISTAAT BAYERN

Table of contents

• 1 About the project
• 2 EDA
  • 2.1 information technology and services indestry’s yearly new recruit count
  • 2.2 freistaat bayern’s workforce status over the years
• 3 Persona company network graph
• 4 Workforce sample
  • 4.1 Achim Mathias
  • 4.2 Alexandra Voll
  • 4.3 Amin Rochdi
  • 4.4 Andreas Eckel
  • 4.5 Armin Friedl
  • 4.6 Birgit Weckerle
  • 4.7 Dark Phoenix
  • 4.8 Edmund Stummer
  • 4.9 Fabian Schwarzfischer
  • 4.10 Florian Schweyer
  • 4.11 Frank Mitzlaff
  • 4.12 Gerrit Bulgrin
  • 4.13 Julian Burmeister-Bießle
  • 4.14 Laura Kahn
  • 4.15 Manfred Dauster
  • 4.16 Marion Schuster
  • 4.17 Mirela Dedic
  • 4.18 Norbert Voit
  • 4.19 Regine Flasche
  • 4.20 Roman Sonnhof
  • 4.21 Sabine Huber
  • 4.22 Samy Sakka
  • 4.23 Suna Yalcin
  • 4.24 Verena Murillo
  • 4.25 Wenzel Steinmetz
  • 4.26 William Kurta
  • 4.27 Winfried Gintschel

## SETUP
from pathlib import Path
import duckdb
from tqdm.notebook import tqdm
import datetime
import copy
import polars as pl
import plotly.express as px
import plotly.io as pio
import re
from concurrent.futures import ThreadPoolExecutor
import plotly.graph_objects as go
import networkx as nx
import numpy as np
# pio.renderers.default = 'plotly_mimetype'
pio.renderers.default = 'jupyterlab+notebook'
pio.templates.default = "plotly_white"

path_data = Path.cwd()/'data'/'03_rdb'
path_data_companies = path_data/'companies_table.parquet'
path_data_experience = path_data/'experience_table.parquet'
path_data_emails = path_data/'emails_table.parquet'
path_data_education = path_data/'education_table.parquet'
path_data_school = path_data/'school_table.parquet'
path_data_persona = path_data/'persona_table.parquet'
path_data_profiles = path_data/'profiles_table.parquet'

path_output_images = Path.cwd()/'output'/'images'

conn = duckdb.connect()

conn.execute("SET temp_directory = 'temp';")
conn.execute("SET memory_limit = '10GB';")
conn.execute("SET max_temp_directory_size = '100GB';")
conn.execute("SET threads = 8;")
conn.execute("SET preserve_insertion_order = false;")
conn.execute("SET enable_progress_bar = true;")
conn.execute("SET enable_progress_bar_print = true;")

df = pl.read_parquet('03_target_companies3.parquet')
df_yearly_new_hires_per_indestry = pl.read_parquet('03_yearly_new_hires_per_indestry.parquet')

current_company_id = "&-friends"
current_company_id = pl.read_json("04__control__.json")[0,'current_company_id']
query = f"""
SELECT *
FROM read_parquet('{path_data_companies}')
WHERE company_id = '{current_company_id}'
"""
df_company_by_company_id = pl.DataFrame(conn.execute(query).df())

current_company_name = df_company_by_company_id[0,'company_name']
current_company_indestry = df_company_by_company_id[0,'company_industry']

current_company_parquet = Path.cwd()/'output'/'company_data'/f"{current_company_id}.parquet"

# Info about personas status from company_id

query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE company_id = '{current_company_id}'
"""
df_experiences_by_company_id = pl.DataFrame(conn.execute(query).df())

personas_whitout_end_date = df_experiences_by_company_id.filter(pl.col('end_date').is_null())
personas_who_got_raise = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) &
                                     pl.col('persona_id').is_in(personas_whitout_end_date['persona_id'].to_list()))
personas_who_stayed = (pl
                      .concat([personas_whitout_end_date, personas_who_got_raise])
                      .sort('start_date')
                      .group_by('persona_id')
                      .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                      )
                      .with_columns(
                          pl.lit(True).alias('still_associated'),
                          pl.lit(None).alias('end_date')
                      )
                      .sort('changes')
                             )

personas_who_left = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) & ~pl.col('persona_id').is_in(personas_who_stayed['persona_id'].to_list()) )
personas_who_left = (personas_who_left
                     .sort('start_date')
                     .group_by('persona_id')
                     .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                              )
                     .with_columns(
                         pl.lit(False).alias('still_associated'),
                         
                     )
                     .sort('changes'))

df_personas_who_worked_in_company = pl.concat([personas_who_stayed, personas_who_left], how='vertical_relaxed').with_columns(
    (pl.col('end_date').dt.year()-pl.col('start_date').dt.year()).alias('work_durration')
).sort('work_durration')

import dns.resolver
import smtplib
import socket

def check_deliverability(email_address):
    """
    Checks the deliverability of an email address by verifying MX records
    and performing an SMTP connection test.
    """
    if '@' not in email_address:
        return False
    
    domain = email_address.split('@')[1]
    
    # Check for MX records
    try:
        mx_records = dns.resolver.resolve(domain, 'MX')
        if not mx_records:
            return False
    except (dns.resolver.NoAnswer, dns.resolver.NXDOMAIN, dns.resolver.Timeout):
        return False

    # Perform SMTP connection test
    mx_host = str(mx_records[0].exchange)
    
    # Validate MX hostname before attempting connection
    try:
        # Test if hostname can be properly encoded
        mx_host.encode('idna')
    except UnicodeError:
        return False
    
    try:
        with smtplib.SMTP(mx_host, timeout=10) as smtp:
            smtp.set_debuglevel(0)
            smtp.helo(socket.gethostname())
            smtp.mail('test@example.com')
            code, _ = smtp.rcpt(email_address)

            return code == 250  # 250 indicates valid email address
            
    except (smtplib.SMTPException, socket.error, UnicodeError):
        return False

# info of all personas info
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_persona}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas = pl.DataFrame(conn.execute(query).df())

# info of all personas profiles
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_profiles}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_profile = pl.DataFrame(conn.execute(query).df())
df_all_personas_profile_f = df_all_personas_profile.group_by('persona_id').agg(pl.col('url').unique())

# info of all personas email
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_emails}')
WHERE persona_id IN ({list_for_in}) AND type == 'personal'
"""
df_all_personas_emails = pl.DataFrame(conn.execute(query).df())

def def_polars_fix_gmail(x):
    if "@gmail" in x:
        first_part = x.split('@')[0]
        second_part = x.split('@')[1]
        return f"{first_part.replace(".",'')}@{second_part}"
    else:
        return x

df_all_personas_emails_f = (df_all_personas_emails
                            .with_columns(pl.col('address')
                                          .map_elements(def_polars_fix_gmail, return_dtype=pl.String)
                                          .alias('normalised_emails'))
                            .unique('normalised_emails', keep='first')
                            .sort('persona_id')
                            .drop('normalised_emails')
                         )
df_all_personas_emails_f = (df_all_personas_emails_f.group_by('persona_id').agg(pl.col('address').unique(),pl.col('type').unique()))
df_all_personas_plus = df_all_personas.join(df_all_personas_emails_f, on='persona_id', how='left')

df_full_personas_who_worked_in_company = (df_personas_who_worked_in_company
                                       .join(df_all_personas_plus, on='persona_id', how='left')
                                       .join(df_all_personas_profile_f, on='persona_id', how='left')
                                      )

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        (pl.col("start_date").fill_null(pl.col("start_date").min()))
        .dt.year()
        .alias("start_year"),
        (pl.col("end_date").dt.year()).alias("end_year"),
    )
)

work_years = []
for i in range(len(df_full_personas_who_worked_in_company)):
    start_y = df_full_personas_who_worked_in_company[i, "start_year"]
    if df_full_personas_who_worked_in_company[i, "end_year"]:
        end_y = df_full_personas_who_worked_in_company[i, "end_year"]
    else:
        end_y = 2020

    tmp_work_years = []
    for y in range(start_y, end_y + 1):
        tmp_work_years.append(y)

    work_years.append(tmp_work_years)

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        pl.Series("work_years", work_years)
    )
)

# add hireups
title_name_match = ["ceo","chief","founder","owner","president","vp","vice","director",
    "cfo","cto","partner","head of","hr ","human","talent","senior","manager","lead"]

df_full_personas_who_worked_in_company = (df_full_personas_who_worked_in_company
    .with_columns(
        pl.when(pl.col('title_name').str.contains_any(title_name_match)).then(True).otherwise(False).alias("higher_up")
    ))



df_tmp_email_checker = (
    df_full_personas_who_worked_in_company
    .filter(
            pl.col('still_associated')==True,
            pl.col('address').list.len()>0
    )
        ['persona_id','address']
        .explode('address')
)

# if current_company_parquet.exists():
#     df_pre_full_personas_who_worked_in_company = pl.read_parquet(current_company_parquet)
#     list_pre_deliverable_address = df_pre_full_personas_who_worked_in_company['address'].drop_nulls().explode().to_list()
# else:
#     list_pre_deliverable_address = []

# list_of_emails_to_check = df_tmp_email_checker['address'].drop_nulls().to_list()
# list_lists_email_check = []

# var_total_emails = len(list_of_emails_to_check)
# var_current_email_count = 0

# def def_check_and_populate(email_to_check):
#     global list_lists_email_check, var_current_email_count
#     if email_to_check in list_pre_deliverable_address:
#         list_lists_email_check.append([email_to_check, True])
#     elif '@gmail' in email_to_check:
#         list_lists_email_check.append([email_to_check, True])
#     else:
#         try:
#             is_deliverable = check_deliverability(email_to_check)
#             list_lists_email_check.append([email_to_check, is_deliverable])
#         except:
#             list_lists_email_check.append([email_to_check, False])
#     var_current_email_count += 1
#     print(' '*10, end='\r')
#     print(round(var_current_email_count/var_total_emails,5), end='\r')

# with ThreadPoolExecutor(max_workers=20) as executor:
#     results = list(executor.map(def_check_and_populate, list_of_emails_to_check))

# df_email_check = pl.DataFrame(list_lists_email_check, schema=["address", "deliverable"], orient="row")
# try:
#     df_tmp_email_checker_f = (
#         df_tmp_email_checker
#             .join(df_email_check, on='address')
#             .filter(pl.col('deliverable')==True)
#             .group_by('persona_id').agg(pl.col('address').unique().alias("deliverable_address"))
#     )
# except:
#     df_tmp_email_checker_f = pl.DataFrame()

# if df_tmp_email_checker_f.is_empty():
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker.rename({'address':'deliverable_address'}), on="persona_id", how='left')
# else:
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker_f, on="persona_id", how='left')

# Info about personas experiences

# info of all experiences[]
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_experiences = pl.DataFrame(conn.execute(query).df())


# info of all comapnies in said experiences
list_w = []
for word in df_all_personas_experiences['company_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT company_id, company_name, company_industry, company_linkedin_url, company_location_country
FROM read_parquet('{path_data_companies}')
WHERE company_id IN ({list_for_in})
"""
df_all_companies = pl.DataFrame(conn.execute(query).df())

df_full_personas_experiences_plus = df_all_personas_experiences.join(df_all_companies, on='company_id', how='left')

df_full_personas_experiences_plus = (
    df_full_personas_experiences_plus
    .with_columns(
        pl.when(
            pl.col('company_id')==current_company_id
        )
        .then(True)
        .otherwise(False)
        .alias('target')
    )
)

# Info about personas education

# info of all experiences
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_education}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_education = pl.DataFrame(conn.execute(query).df())


#ifon of allcomapnies in said experiences
list_w = []
for word in df_all_personas_education['school_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

if list_w:
    list_for_in = ', '.join(list_w)
    query = f"""
    SELECT school_id, school_name, school_type, school_website, school_location_country
    FROM read_parquet('{path_data_school}')
    WHERE school_id IN ({list_for_in})
    """
    df_all_school = pl.DataFrame(conn.execute(query).df())
    
    df_full_personas_education_plus = df_all_personas_education.join(df_all_school, on='school_id', how='left')
else:
    df_full_personas_education_plus = df_all_personas_education

1 About the project

The project came to life after realizing that web scraping doesn’t allow deep-level filtering—without consuming too much time.The irony is, this project itself took me about a month, but the final RDB contains more data than I could ever scrape.

The raw data was 1.4 TB in size and holds information previously scraped.
Processing was done on my local machine using Python, Polars, and DuckDB, following this workflow:
- Processed raw data into structured Parquet files using Polars.
- Transformed each Parquet file into mini RDBs using Polars.
- Merged all mini RDBs into one using DuckDB.
- Analyzed and filtered data to fit the current project.

2 EDA

2.1 information technology and services indestry’s yearly new recruit count

list_of_unique_company_experience_years = []
for y in df_full_personas_who_worked_in_company['start_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)
for y in df_full_personas_who_worked_in_company['end_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)

list_year = []
list_state = []
list_count = []
list_names = []

def def_get_names_breked(tmp):
    if tmp.is_empty():
        names_string = ''
    else:
        tmp_list_name = []
        names_limit = 3
        row_limit = names_limit * 6
        for i, name in enumerate(tmp['full_name'].to_list()):
            ii = i+1
            tmp_list_name.append(name.title())
            if ii!=0 and ii%names_limit==0:
                tmp_list_name.append("<br>")
            if ii==row_limit:
                tmp_list_name.append("...")
                break
        names_string = ', '.join(tmp_list_name).replace(", <br>, ","<br>")
    return names_string

for y in list_of_unique_company_experience_years:
    #recuite state
    list_year.append(y)
    list_state.append('Recruited')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('start_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))
    
    #recuite state
    list_year.append(y)
    list_state.append('Resigned')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('end_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))

df_m_recruite_vs_resign = pl.DataFrame({
    'year':list_year,
    'status':list_state,
    'count':list_count,
    'names':list_names,})

2.2 freistaat bayern’s workforce status over the years

3 Persona company network graph

gr_net = df_full_personas_experiences_plus.with_columns(pl.col('company_id').str.to_uppercase()).group_by('persona_id','company_id').agg(pl.len().alias('count')).sort('count')
list_top_in_network = gr_net['company_id'].value_counts().sort('count', descending=True)['company_id'].to_list()[:5]
gr_net_f = gr_net.filter(pl.col('company_id').is_in(list_top_in_network))

list_letters = ['A','B','C','D','E','F','G','H']
dict_company = {}
dict_company_rev = {}
for company, letter in zip(list_top_in_network, list_letters ):
    dict_company[letter] = company
    dict_company_rev[company] = letter

gr_gr_net_f = gr_net_f.sort('company_id').group_by('persona_id').agg(pl.col('company_id').unique().sort(),)

gr_gr_net_f2 = (
    gr_gr_net_f['company_id']
    .value_counts()
    .with_columns(
        # pl.col('company_id').list.join(', '),
        (pl.col('count')/len(gr_gr_net_f)).alias('per')
    )
    .sort('per',descending=True)
)

list_prob = []
for i in range(len(gr_gr_net_f2)):
    tmp_prob_letters = []
    for k in dict_company.keys():
        if dict_company[k] in gr_gr_net_f2[i]['company_id'][0].to_list():
            tmp_prob_letters.append(f' {k}')
        else:
            tmp_prob_letters.append(f'¬{k}')

    list_prob.append(f"P({' ∩ '.join(tmp_prob_letters)}) = {round(gr_gr_net_f2[i]['per'][0],4)}")
annon_prob_text = "<b>Probability Distribution:</b><br>" + '<br>'.join(list_prob)



# Create network graph
G = nx.Graph()
for persona, company in gr_net_f.select(['persona_id', 'company_id']).iter_rows():
    G.add_edge(persona, company)

# Get unique values
persona_ids = gr_net_f['persona_id'].unique().to_list()
company_ids = gr_net_f['company_id'].unique().to_list()

# Calculate degrees (connection counts)
degree_dict = dict(G.degree())

# Get min and max degrees for scaling
company_degrees = [degree_dict[c] for c in company_ids]
persona_degrees = [degree_dict[p] for p in persona_ids]

min_company_degree = min(company_degrees) if company_degrees else 1
max_company_degree = max(company_degrees) if company_degrees else 1
min_persona_degree = min(persona_degrees) if persona_degrees else 1
max_persona_degree = max(persona_degrees) if persona_degrees else 1

# Define size ranges
COMPANY_MIN_SIZE = 25
COMPANY_MAX_SIZE = 100
PERSONA_MIN_SIZE = 5
PERSONA_MAX_SIZE = 20

# print(f"Company connections range: {min_company_degree} - {max_company_degree}")
# print(f"Persona connections range: {min_persona_degree} - {max_persona_degree}")

# Sort companies by degree (size) in descending order
company_ids_sorted = sorted(company_ids, key=lambda x: degree_dict[x], reverse=True)

# Check if "Nokia" exists in the data
HIGHLIGHTED_COMPANY = current_company_id
HIGHLIGHTED_COMPANY_EXISTS = HIGHLIGHTED_COMPANY.lower() in [str(c).lower() for c in company_ids]

if HIGHLIGHTED_COMPANY_EXISTS:
    # Get the actual case-sensitive name
    highlighted_company = next(c for c in company_ids if str(c).lower() == HIGHLIGHTED_COMPANY.lower())
    # print(f"Highlighting company: {highlighted_company} (with {degree_dict[highlighted_company]} connections)")
else:
    # print(f"Warning: '{HIGHLIGHTED_COMPANY}' not found in company list")
    highlighted_company = None

# Create layout (companies on outer circle, ordered by size)
pos = {}
num_companies = len(company_ids_sorted)
radius_outer = 2.0

# Position companies on circle, ordered by size (largest first)
for i, company in enumerate(company_ids_sorted):
    # Start at top (90° or π/2 radians) and go counter-clockwise (add angle)
    # Counter-clockwise rotation: angle = start_angle + (i * 2π / num_companies)
    # This puts largest at top, next on left, then bottom, then right
    start_angle = np.pi / 2  # 90° at top
    
    # For counter-clockwise rotation
    angle = start_angle - (2 * np.pi * i / num_companies)
    
    # Convert to x, y coordinates
    pos[company] = (radius_outer * np.cos(angle), radius_outer * np.sin(angle))

# Position personas
for i, persona in enumerate(persona_ids):
    connected_companies = [c for c in company_ids if G.has_edge(persona, c)]
    if connected_companies:
        avg_x = np.mean([pos[c][0] for c in connected_companies])
        avg_y = np.mean([pos[c][1] for c in connected_companies])
        # Add jitter to spread out personas
        jitter_x = np.random.uniform(-0.2, 0.2)
        jitter_y = np.random.uniform(-0.2, 0.2)
        pos[persona] = (avg_x * 0.5 + jitter_x, avg_y * 0.5 + jitter_y)
    else:
        pos[persona] = (0, 0)

# Prepare edge traces
edge_x, edge_y = [], []
for edge in G.edges():
    x0, y0 = pos[edge[0]]
    x1, y1 = pos[edge[1]]
    edge_x.extend([x0, x1, None])
    edge_y.extend([y0, y1, None])

edge_trace = go.Scatter(
    x=edge_x, y=edge_y,
    line=dict(width=0.6, color='rgba(120, 120, 120, 0.15)'),
    hoverinfo='none',
    mode='lines')

# Prepare node traces with proportional sizing
company_x, company_y, company_text = [], [], []
company_color, company_size, company_hover = [], [], []
company_border_width = []  # For border thickness
company_border_color = []  # For border color

persona_x, persona_y = [], []
persona_color, persona_size, persona_hover = [], [], []

# Helper function to scale size proportionally
def scale_size(value, min_val, max_val, min_size, max_size):
    if max_val == min_val:
        return (min_size + max_size) / 2
    return min_size + (value - min_val) / (max_val - min_val) * (max_size - min_size)

# Add COMPANY nodes in sorted order (largest first)
for company in company_ids_sorted:
    x, y = pos[company]
    company_x.append(x)
    company_y.append(y)
    company_text.append(str(company))
    company_color.append('#EF553B')
    
    connections = degree_dict[company]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections, 
        min_company_degree, 
        max_company_degree,
        COMPANY_MIN_SIZE, 
        COMPANY_MAX_SIZE
    )
    company_size.append(scaled_size)
    
    # Custom border for highlighted company
    if highlighted_company and company == highlighted_company:
        company_border_width.append(4)  # Thicker border
        company_border_color.append('#000000')  # Black border
    else:
        company_border_width.append(1)
        company_border_color.append('#000000')
    
    # Hover text
    personas = gr_net_f.filter(pl.col('company_id') == company)['persona_id'].to_list()
    rank = company_ids_sorted.index(company) + 1
    hover_text = f"<b>Company #{rank}:</b> {company}<br>"
    hover_text += f"<b>Personas worked here:</b> {connections}<br>"
    hover_text += f"<b>Connection rank:</b> {rank}/{len(company_ids_sorted)}<br>"
    if connections > 0:
        for persona in personas[:5]:
            persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
            hover_text += f" • {persona_name}<br>"
        if connections > 5:
            hover_text += f" • ... and {connections - 5} more"
    company_hover.append(hover_text)

# Add PERSONA nodes
for persona in persona_ids:
    x, y = pos[persona]
    persona_x.append(x)
    persona_y.append(y)
    persona_color.append('#636efa')
    
    connections = degree_dict[persona]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections,
        min_persona_degree,
        max_persona_degree,
        PERSONA_MIN_SIZE,
        PERSONA_MAX_SIZE
    )
    persona_size.append(scaled_size)
    
    # Hover text
    companies = gr_net_f.filter(pl.col('persona_id') == persona)['company_id'].to_list()
    persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
    hover_text = f"<b>Persona:</b> {persona_name}<br>"
    hover_text += f"<b>Companies worked at:</b> {connections}<br>"
    if connections > 0:
        # Check if worked at highlighted company
        if highlighted_company:
            worked_at_highlighted = highlighted_company in companies
            if worked_at_highlighted:
                hover_text += f"<b>Worked at {highlighted_company}:</b> ✓<br>"
        
        hover_text += "<br>".join([f"  • {comp}" for comp in companies[:5]])
        if connections > 5:
            hover_text += f"<br>  • ... and {connections - 5} more"
    persona_hover.append(hover_text)

# Create company node trace
company_trace = go.Scatter(
    x=company_x, y=company_y,
    mode='markers+text',
    hoverinfo='text',
    hovertext=company_hover,
    text=company_text,
    textposition="top center",
    textfont=dict(size=14, color='black'),
    marker=dict(
        color=company_color,
        size=company_size,
        line=dict(
            width=company_border_width,
            color=company_border_color
        ),
        opacity=0.9)
)

# Create persona node trace
persona_trace = go.Scatter(
    x=persona_x, y=persona_y,
    mode='markers',
    hoverinfo='text',
    hovertext=persona_hover,
    text=None,  # No text for personas
    marker=dict(
        color=persona_color,
        size=persona_size,
        line=dict(width=1, color='black'),
        opacity=0.7)
)

# Calculate axis ranges for 1:1 aspect ratio
all_positions = list(pos.values())
x_vals = [p[0] for p in all_positions]
y_vals = [p[1] for p in all_positions]

# Add padding
x_range = [min(x_vals) - 0.5, max(x_vals) + 0.5]
y_range = [min(y_vals) - 0.5, max(y_vals) + 0.5]

# Make axes have the same range for 1:1 aspect
max_range = max(x_range[1] - x_range[0], y_range[1] - y_range[0])
x_center = (x_range[0] + x_range[1]) / 2
y_center = (y_range[0] + y_range[1]) / 2

x_range = [x_center - max_range/2, x_center + max_range/2]
y_range = [y_center - max_range/2, y_center + max_range/2]

# Create figure with 1:1 aspect ratio
fig = go.Figure(data=[edge_trace, persona_trace, company_trace],
                layout=go.Layout(
                    title=f'Persona-Company Network (Companies Ordered by Size)<br><sup>Highlighted: {highlighted_company if highlighted_company else "None"}</sup>',
                    showlegend=False,
                    hovermode='closest',
                    margin=dict(b=20, l=20, r=20, t=100),
                    xaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=x_range,
                        scaleanchor="y",
                        scaleratio=1
                    ),
                    yaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=y_range
                    ),
                    plot_bgcolor='white',
                    paper_bgcolor='white',
                    width=900,
                    height=900
                ))

# Add legend with size examples and highlighting info
# legend_text = f"""
# <b>Node Size = Connection Count</b><br>
# <span style='color:#EF553B'>● Companies</span><br>
# <span style='color:#636efa'>● Personas</span> (hover for details)
# """

# fig.add_annotation(
#     x=0.98, y=0.98,
#     xref="paper", yref="paper",
#     text=legend_text,
#     showarrow=False,
#     font=dict(size=14),
#     align="left",
#     bgcolor="rgba(255, 255, 255, 0.95)",
    
# )

# Add top companies list
top_companies = company_ids_sorted[:10]  # Top 10 companies
top_companies_text = "<b>Top Companies by Connections:</b><br>"
for i, company in enumerate(top_companies, 1):
    connections = degree_dict[company]
    top_connections = degree_dict[top_companies[0]]
    connections_per = f" | {round(connections/top_connections*100)}%" if highlighted_company and company != highlighted_company else ""
    highlight_indicator = " " if highlighted_company and company == highlighted_company else ""
    top_companies_text += f"{dict_company_rev[company]}. {company}: {connections} {connections_per} {highlight_indicator}<br>"

fig.add_annotation(
    x=0.02, y=0.98,
    xref="paper", yref="paper",
    text=top_companies_text,
    showarrow=False,
    font=dict(size=14),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.9)",
    # bordercolor="#666",
    # borderwidth=1
)

# Add probabiliy list

fig.add_annotation(
    x=0.98, y=0.98,
    xref="paper", yref="paper",
    text=annon_prob_text,
    showarrow=False,
    font=dict(
        family="'Courier New', monospace",  # Multiple fallbacks
        size=12,
        color="black"
    ),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.95)",
    
)
fig.write_image((path_output_images/f'network_{current_company_id}.webp'))
fig.show()

amount = 5

tmp = df_full_personas_who_worked_in_company.sort(
    ["inferred_salary", "linkedin_connections", "inferred_years_experience"],
    descending=True,
)
tmp_gr = df_full_personas_experiences_plus.group_by('persona_id').agg(pl.len().alias('experience_count'))
tmp = df_full_personas_who_worked_in_company.join(tmp_gr, on='persona_id').sort('experience_count',descending=True)

tmp2 = pl.concat(
    [tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==True)[:amount*2],
     tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==False)[:amount*2],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==True)[:amount*1],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==False)[:amount*1],
    ]
).sort("full_name")

list_persona_for_plot = tmp2['persona_id'].to_list()

# Workforce data

def def_plotly_experience_range(current_persona_id):
    tmp_df = (df_full_personas_who_worked_in_company
              .filter(pl.col('persona_id')==current_persona_id)
              .with_columns(pl.col('end_year').fill_null(2021))['start_year','end_year'])
    
    fig_tmp = copy.deepcopy(fig_company_hiring_trend)
    fig_tmp.add_vrect(
        x0=tmp_df[0,'start_year'],
        x1=tmp_df[0,'end_year'],
        fillcolor="blue",
        opacity=0.1,
        line_width=0 
    )
    return fig_tmp

def def_plotly_experience_gantt(current_persona_id):
    px_data = (df_full_personas_experiences_plus
               .filter(pl.col('persona_id')==current_persona_id)
               .with_columns(
                   pl.col('end_date').fill_null(datetime.datetime(2020, 1, 1, 0,0)),
                   pl.col('company_name').str.to_uppercase(),
                   # pl.col('company_name').str.to_uppercase().str.replace_all('&', '-and-')
               )
               .sort('start_date'))
    
    y_order = px_data['company_name'].to_list()
    
    fig = px.timeline(px_data,x_start="start_date", x_end="end_date", y="company_name",
                      color='target',hover_data=["title_name"], height=140+30*len(px_data),
                      category_orders={"company_name": y_order},
                      color_discrete_map={True:'#EF553B',  False:'#636efa'},
                      labels={'target':'Target', 'start_date':'Recruited', 'end_date':'If-Resigned', 
                             'company_name':'Company', 'title_name':'Job role'}
                     # title=f"Experience of {current_persona_name}.",
                     )
    fig.update_yaxes(
        # autorange="reversed",
                              showgrid=True,
                              gridcolor='lightgray',
                              gridwidth=1,
                              griddash='dot'
    )
    fig.update_layout(showlegend=False, xaxis_title=None, yaxis_title=None)
    return fig

4 Workforce sample

4.1 Achim Mathias

Job title: Senior-consultant
Socials: https://linkedin.com/in/achim-mathias-797819a9

4.1.1 Achim Mathias’s working period at freistaat bayern

4.1.2 Gantt plot of Achim Mathias’s experience

---

4.2 Alexandra Voll

Job title: Instructor
Socials: https://linkedin.com/in/alexandra-voll-921052b5

4.2.1 Alexandra Voll’s working period at freistaat bayern

4.2.2 Gantt plot of Alexandra Voll’s experience

---

4.3 Amin Rochdi

Job title: Studienrat im realschuldienst
Socials: https://linkedin.com/in/amin-rochdi-1a09141b | https://facebook.com/amin.rochdi.96

4.3.1 Amin Rochdi’s working period at freistaat bayern

4.3.2 Gantt plot of Amin Rochdi’s experience

---

4.4 Andreas Eckel

Job title: Official in the public service
Socials: https://linkedin.com/in/andreas-eckel-767019177

4.4.1 Andreas Eckel’s working period at freistaat bayern

4.4.2 Gantt plot of Andreas Eckel’s experience

---

4.5 Armin Friedl

Job title: Wirtschaftspã dagoge
Socials: https://linkedin.com/in/armin-friedl-a66b7582

4.5.1 Armin Friedl’s working period at freistaat bayern

4.5.2 Gantt plot of Armin Friedl’s experience

---

4.6 Birgit Weckerle

Job title: Studienreferendarin
Socials: https://linkedin.com/in/birgit-weckerle-b1324180

4.6.1 Birgit Weckerle’s working period at freistaat bayern

4.6.2 Gantt plot of Birgit Weckerle’s experience

---

4.7 Dark Phoenix

Job title: Chief executive officer
Socials: https://linkedin.com/in/dark-phoenix-29012219a

4.7.1 Dark Phoenix’s working period at freistaat bayern

4.7.2 Gantt plot of Dark Phoenix’s experience

---

4.8 Edmund Stummer

Job title: Senior auditor
Socials: https://linkedin.com/in/edmund-stummer-2952aa78

4.8.1 Edmund Stummer’s working period at freistaat bayern

4.8.2 Gantt plot of Edmund Stummer’s experience

---

4.9 Fabian Schwarzfischer

Job title: Rechtsreferendar
Socials: https://linkedin.com/in/schwarzfischer

4.9.1 Fabian Schwarzfischer’s working period at freistaat bayern

4.9.2 Gantt plot of Fabian Schwarzfischer’s experience

---

4.10 Florian Schweyer

Job title: Richter am landgericht
Socials: https://linkedin.com/in/dr-florian-schweyer-ll-m-7609837

4.10.1 Florian Schweyer’s working period at freistaat bayern

4.10.2 Gantt plot of Florian Schweyer’s experience

---

4.11 Frank Mitzlaff

Job title: Servicetechniker und verwaltungsfachwirt
Socials: https://linkedin.com/in/frank-mitzlaff-0437a293

4.11.1 Frank Mitzlaff’s working period at freistaat bayern

4.11.2 Gantt plot of Frank Mitzlaff’s experience

---

4.12 Gerrit Bulgrin

Job title: Notarassessor
Socials: https://linkedin.com/in/dr-gerrit-bulgrin-26348473 | https://linkedin.com/in/gerrit-bulgrin-26348473

4.12.1 Gerrit Bulgrin’s working period at freistaat bayern

4.12.2 Gantt plot of Gerrit Bulgrin’s experience

---

4.13 Julian Burmeister-Bießle

Job title: Richter am arbeitsgericht
Socials: https://linkedin.com/in/dr-julian-raphael-burmeister-bießle-b609a5ba

4.13.1 Julian Burmeister-Bießle’s working period at freistaat bayern

4.13.2 Gantt plot of Julian Burmeister-Bießle’s experience

---

4.14 Laura Kahn

Job title: Account-manager
Socials: https://linkedin.com/in/laura-kahn-251499149

4.14.1 Laura Kahn’s working period at freistaat bayern

4.14.2 Gantt plot of Laura Kahn’s experience

---

4.15 Manfred Dauster

Job title: Vorsitzender richter am oberlandesgericht
Socials: https://linkedin.com/in/manfred-dr-dauster-a9824636

4.15.1 Manfred Dauster’s working period at freistaat bayern

4.15.2 Gantt plot of Manfred Dauster’s experience

---

4.16 Marion Schuster

Job title: Trainee lawyer
Socials: https://linkedin.com/in/marion-schuster-07097722

4.16.1 Marion Schuster’s working period at freistaat bayern

4.16.2 Gantt plot of Marion Schuster’s experience

---

4.17 Mirela Dedic

Job title: Manager
Socials: https://linkedin.com/in/mirela-dedic-9139821ba

4.17.1 Mirela Dedic’s working period at freistaat bayern

4.17.2 Gantt plot of Mirela Dedic’s experience

---

4.18 Norbert Voit

Job title: Senior tax auditor
Socials: https://linkedin.com/in/norbert-voit-12402312b

4.18.1 Norbert Voit’s working period at freistaat bayern

4.18.2 Gantt plot of Norbert Voit’s experience

---

4.19 Regine Flasche

Job title: Art instructor
Socials: https://linkedin.com/in/regine-flasche-1196341b0

4.19.1 Regine Flasche’s working period at freistaat bayern

4.19.2 Gantt plot of Regine Flasche’s experience

---

4.20 Roman Sonnhof

Job title: Manager beamter
Socials: https://linkedin.com/in/roman-sonnhof-aabb03198

4.20.1 Roman Sonnhof’s working period at freistaat bayern

4.20.2 Gantt plot of Roman Sonnhof’s experience

---

4.21 Sabine Huber

Job title: Referendar
Socials: https://linkedin.com/in/sabine-huber-b2a101110 | https://linkedin.com/in/sabine-susanne-huber

4.21.1 Sabine Huber’s working period at freistaat bayern

4.21.2 Gantt plot of Sabine Huber’s experience

---

4.22 Samy Sakka

Job title: Rechtsreferendar
Socials: https://linkedin.com/in/samy-sakka-a1a65a74 | https://linkedin.com/in/samysakka

4.22.1 Samy Sakka’s working period at freistaat bayern

4.22.2 Gantt plot of Samy Sakka’s experience

---

4.23 Suna Yalcin

Job title: Rechtsreferendarin
Socials: https://linkedin.com/in/suna-yalcin-07359447

4.23.1 Suna Yalcin’s working period at freistaat bayern

4.23.2 Gantt plot of Suna Yalcin’s experience

---

4.24 Verena Murillo

Job title: Secondary school teacher for english and geography
Socials: https://linkedin.com/in/verena-rivas-murillo-70715468

4.24.1 Verena Murillo’s working period at freistaat bayern

4.24.2 Gantt plot of Verena Murillo’s experience

---

4.25 Wenzel Steinmetz

Job title: Notarassessor
Socials: https://linkedin.com/in/wenzel-steinmetz-406695120

4.25.1 Wenzel Steinmetz’s working period at freistaat bayern

4.25.2 Gantt plot of Wenzel Steinmetz’s experience

---

4.26 William Kurta

Job title: Civil service
Socials: https://linkedin.com/in/william091

4.26.1 William Kurta’s working period at freistaat bayern

4.26.2 Gantt plot of William Kurta’s experience

---

4.27 Winfried Gintschel

Job title: Schulleiter
Socials: https://linkedin.com/in/winfried-gintschel-aa2a1991 | https://facebook.com/winfried.gintschel

4.27.1 Winfried Gintschel’s working period at freistaat bayern

4.27.2 Gantt plot of Winfried Gintschel’s experience

---

df_full_personas_who_worked_in_company.write_parquet(current_company_parquet)